Training Set Research Articles

SFNet: Spatial and Frequency Domain Networks for Wide-Field OCT Angiography Retinal Vessel Segmentation.

Automatic segmentation of blood vessels in fundus images is important to assist ophthalmologists in diagnosis. However, automatic segmentation for Optical Coherence Tomography Angiography (OCTA) blood vessels has not been fully investigated due to various difficulties, such as vessel complexity. In addition, there are only a few publicly available OCTA image data sets for training and validating segmentation algorithms. To address these issues, we constructed a wild-field retinal OCTA segmentation data set, the Retinal Vessels Images in OCTA (REVIO) dataset. Second, we propose a new retinal vessel segmentation network based on spatial and frequency domain networks (SFNet). The proposed model are tested on three benchmark data sets including REVIO, ROSE and OCTA-500. The experimental results show superior performance on segmentation tasks compared to the representative methods.

Explainable machine learning model for predicting the risk of significant liver fibrosis in patients with diabetic retinopathy.

Diabetic retinopathy (DR), a prevalent complication in patients with type 2 diabetes, has attracted increasing attention. Recent studies have explored a plausible association between retinopathy and significant liver fibrosis. The aim of this investigation was to develop a sophisticated machine learning (ML) model, leveraging comprehensive clinical datasets, to forecast the likelihood of significant liver fibrosis in patients with retinopathy and to interpret the ML model by applying the SHapley Additive exPlanations (SHAP) method. This inquiry was based on data from the National Health and Nutrition Examination Survey 2005-2008 cohort. Utilizing the Fibrosis-4 index (FIB-4), liver fibrosis was stratified across a spectrum of grades (F0-F4). The severity of retinopathy was determined using retinal imaging and segmented into four discrete gradations. A ten-fold cross-validation approach was used to gauge the propensity towards liver fibrosis. Eight ML methodologies were used: Extreme Gradient Boosting, Random Forest, multilayer perceptron, Support Vector Machines, Logistic Regression (LR), Plain Bayes, Decision Tree, and k-nearest neighbors. The efficacy of these models was gauged using metrics, such as the area under the curve (AUC). The SHAP method was deployed to unravel the intricacies of feature importance and explicate the inner workings of the ML model. The analysis included 5,364 participants, of whom 2,116 (39.45%) exhibited notable liver fibrosis. Following random allocation, 3,754 individuals were assigned to the training set and 1,610 were allocated to the validation cohort. Nine variables were curated for integration into the ML model. Among the eight ML models scrutinized, the LR model attained zenith in both AUC (0.867, 95% CI: 0.855-0.878) and F1 score (0.749, 95% CI: 0.732-0.767). In internal validation, this model sustained its superiority, with an AUC of 0.850 and an F1 score of 0.736, surpassing all other ML models. The SHAP methodology unveils the foremost factors through importance ranking. Sophisticated ML models were crafted using clinical data to discern the propensity for significant liver fibrosis in patients with retinopathy and to intervene early. Improved early detection of liver fibrosis risk in retinopathy patients enhances clinical intervention outcomes.

Reliability of plastid and mitochondrial localisation prediction declines rapidly with the evolutionary distance to the training set increasing.

Mitochondria and plastids import thousands of proteins. Their experimental localisation remains a frequent task, but can be resource-intensive and sometimes impossible. Hence, hundreds of studies make use of algorithms that predict a localisation based on a protein's sequence. Their reliability across evolutionary diverse species is unknown. Here, we evaluate the performance of common algorithms (TargetP, Localizer and WoLFPSORT) for four photosynthetic eukaryotes (Arabidopsis thaliana, Zea mays, Physcomitrium patens, and Chlamydomonas reinhardtii) for which experimental plastid and mitochondrial proteome data is available, and 171 eukaryotes using orthology inferences. The match between predictions and experimental data ranges from 75% to as low as 2%. Results worsen as the evolutionary distance between training and query species increases, especially for plant mitochondria for which performance borders on random sampling. Specificity, sensitivity and precision analyses highlight cross-organelle errors and uncover the evolutionary divergence of organelles as the main driver of current performance issues. The results encourage to train the next generation of neural networks on an evolutionary more diverse set of organelle proteins for optimizing performance and reliability.

Efficient Cross-Domain Fault Diagnosis via Distributed Multi-Source Domain Deep Transfer Learning

Abstract In the actual industrial production, the working conditions of rotating machinery are complex and changeable, and the health-state monitoring data are increasingly large and difficult to be labeled, which will seriously restrict the accuracy and efficiency of cross-domain fault diagnosis (CDFD) of rotating machinery. Therefore, an efficient multi-source domain deep transfer learning (MDDTL) method for CDFD of rotating machinery is proposed. Firstly, a MDDTL model is constructed to improve the accuracy of CDFD. In the model, a dual-phase domain alignment strategy is designed, which considers the alignment of feature distributions between each source and target domain pair in the feature space and that of prediction probabilities between domain-specific fault classifiers in the output space. The fault prediction results from multiple different fault classifiers are merged dynamically by the proposed imbalanced adaptive prediction strategy. Secondly, a data-parallel distributed training scheme for MDDTL model is proposed. Based on the idea of data parallelism, the distributed parallel training of MDDTL model is performed with Horovod-GPU platform, and the parameters are synchronously updated with the bandwidth-optimal Ring-AllReduce architecture. Under the premise of ensuring the accuracy of fault diagnosis (FD), the training time of MDDTL model is significantly reduced. Finally, extensive experiments are conducted to verify the effectiveness of the proposed MDDTL method. The results demonstrate that the proposed method not only effectively improves the accuracy of CDFD of rotating machinery, but also significantly improves the training efficiency of MDDTL model. After adopting the proposed method, the diagnosis accuracies achieved under two different cross-working condition scenarios reach 97.09% and 97.87% respectively, and the model training time is reduced by 73.62% when facing a large-scale rotating machinery training set.

NIMG-18. RAPID INTRAOPERATIVE DIFFERENTIATION OF PRIMARY CNS LYMPHOMA FROM COMMON CNS TUMORS USING STIMULATED RAMAN HISTOLOGY AND DEEP LEARNING

Abstract Accurate intraoperative diagnosis is crucial for decision-making during tumor surgery and stereotactic-guided biopsies. Differentiating between primary CNS lymphoma (PCNSL) and non-PCNSL entities such as gliomas and metastatic cancers presents significant challenges due to overlapping histomorphological features, time constraints, and the differing further therapy strategies required. Our study utilized deep learning and stimulated Raman histology (SRH) to address this challenge. We imaged unprocessed, label-free tissue samples intraoperatively using a portable Raman scattering microscope, generating virtual H&E-like images within less than five minutes. Our training set comprised over 54,000 SRH tumor patch images, including various brain tumors. We developed a deep learning pipeline utilizing a ResNet-50 feature extractor and the BYOL (Bootstrap Your Own Latent) self-supervised learning algorithm. We trained a binary linear classifier to distinguish between PCNSL and non-PCNSL entities. Data were collected from New York University (NYU), the University of Cologne (UKK), and the University of Michigan (UM), including 263 SRH whole slide images from 177 patients sourced from surgical resections and stereotactic-guided biopsies. In an international bicentric test cohort (N = 100, NYU and UKK), the model achieved an overall accuracy of 91.1%, an area under the receiver operating characteristic curve (AUROC) of 0.983, a sensitivity of 89.3%, and a specificity of 100% compared to the final formalin-fixed, paraffin-embedded-based diagnosis. Our algorithm can accurately extract typical histomorphological features to help differentiate between PCNSL and non-PCNSL tumors. These results underscore the potential of deep learning models to enhance the accuracy of intraoperative diagnoses for PCNSL and non-PCNSL entities, thereby improving intraoperative decision-making and further treatment strategies by supporting neuropathologists and surgeons.

A Convolution Auto-Encoders Network for Aero-Engine Hot Jet FT-IR Spectrum Feature Extraction and Classification

Aiming at classification and recognition of aero-engines, two telemetry Fourier transform infrared (FT-IR) spectrometers are utilized to measure the infrared spectrum of the areo-engine hot jet, meanwhile a spectrum dataset of six types of areo-engines is established. In this paper, a convolutional autoencoder (CAE) is designed for spectral feature extraction and classification, which is composed of coding network, decoding network, and classification network. The encoder network consists of convolutional layers and maximum pooling layers, the decoder network consists of up-sampling layers and deconvolution layers, and the classification network consists of a flattened layer and a dense layer. In the experiment, data for the spectral dataset were randomly sampled at a ratio of 8:1:1 to produce the training set, validation set, and prediction set, and the performance measures were accuracy, precision, recall, confusion matrix, F1 score, ROC curve, and AUC value. The experimental result of CAE reached 96% accuracy and the prediction running time was 1.57 s. Compared with the classical PCA feature extraction and SVM, XGBoost, AdaBoost, and Random Forest classifier algorithms, as well as AE, CSAE, and CVAE deep learning classification methods, the CAE network can achieve higher accuracy and efficiency and can complete the spectral classification task.

Fault Diagnosis Model of Hydraulic Motor Based on Fuzzy Neural Network

When the hydraulic motor fault occurs,it is not easy to be detected,and the leakage degree will gradually increase.In order toavoid bigger accidents causedby thehydraulic motor fault,the accident isexcluded in theembryonic stage,and the hydraulic motor fault prediction method based onfuzzy neural network isused topredict thehydraulic motor fault.The feature vector is output inthe global meanpooling layer,andthe feature vectormatrix between thehealth state feature vector library and the samples to be measured is constructed.The dynamic cluster graph isobtained by fuzzy clustering,so as to realize the fault diagnosis of thehydraulic motor.The results show that the accuracy of training set,verification set andtest set ishigher than99.8%.The accuracy of diagnosis classification is99.00%,which is better than othercomparison models.In this study,the number of training samples can be appropriately increased ordecreased according to thecurve complexity of the detection target,so as to improve thefeature extraction capability of the convolutional layer andincrease the classification accuracy.

A New Classification Model Using a Decision Tree Generated from Hyperplanes in Dimensional Space

ABSTRACT One of the issues addressed by machine learning, with applications in various disciplines or fields such as the health sector and the agricultural sector among others, involves data classification. For this purpose, various models within supervised learning have been proposed and developed that allow for the classification of these data. However, one of the implications of the No-Free-Lunch theorems is that there is no optimal general-purpose model, i.e. there is no classifier model that achieves the best results for all problems presented. Hence the importance of proposing and implementing new classifier models, evaluating their performance, and comparing them with other classifier models in order to achieve models with good results in specific problems. This work presents a new classifier model that, by constructing hyperplanes from the training set, generates a decision tree and partitions the dimensional space. The proposed model was applied to different problems such as the XOR logical function problem, where the proposed model managed to solve the problem, it was also applied to the Iris Dataset where one of the trees generated by the model managed to classify with 100% accuracy the test set, and finally, the proposed model was applied to the Pima Indians Diabetes Database and compared with other models using the accuracy value. The proposed model obtained an accuracy of 81.81%, achieving the best result in the same way as the Random Forest Classifier. The results obtained in this work show that the proposed model manages to partition the dimensional space adequately from the training set and thus competitively classify the data with other state-of-the-art models.

Forecasting Chinese Stock Market Volatility With Volatilities in Bond Markets

ABSTRACTIn this paper, we investigate whether the bond markets contain important information that can improve the accuracy of stock market volatility forecasts in China. We use realized volatility (RV) implemented by different maturity treasury bond futures contracts to predict the Chinese stock market volatility. Our work is based on the heterogeneous autoregressive (HAR) framework. Empirical results show that the volatility of treasury bond contracts with longer maturities (especially 10 years) has the best effect on predicting the Chinese stock market volatility, both in sample and out of sample. Two machine learning methods, the scaled principal component analysis (SPCA) and the least absolute shrinkage and selection operator (lasso), are also more effective than the HAR benchmark model's prediction. Finally, mean–variance investors can achieve substantial economic gains by allocating their investment portfolios based on volatility forecasts after introducing treasury bond futures volatility.

Bayesian Learning Aided Theoretical Optimization of IrPdPtRhRu High Entropy Alloy Catalysts for the Hydrogen Evolution Reaction.

The complex compositional space of high entropy alloys (HEAs) has shown a great potential to reduce the cost and further increase the catalytic activity for hydrogen evolution reaction (HER) by compositional optimization. Without uncovering the specifics of the HER mechanism on a given HEA surface, it is unfeasible to apply compositional modifications to enhance the performance and save costs. In this work, a combination of density functional theory and Bayesian machine learning is used to demonstrate the unique catalytic mechanism of IrPdPtRhRu HEA catalysts for HER. At high coverage of underpotential-deposited hydrogen, a d-band investigation of the active sites of the HEA surface is conducted to elucidate the superior catalytic performance through electronic interactions between elements. At low coverage, a novel Bayesian learning with oversampling approach is then outlined to optimize the HEA composition for performance improvement and cost reduction. This approach proves more efficacious and efficient and yields higher-quality structures with less training set bias compared with neural-network optimization. The proposed HEA optimization theoretically outperforms benchmark Pt catalysts' overpotential by ≈40% at a 15% reduced synthesis cost comparing to the equiatomic ratio HEA.

Performance Predictions of Solar-Assisted Heat Pumps: Methodological Approach and Comparison Between Various Artificial Intelligence Methods

The coefficient of performance (COP) is a crucial metric for evaluating the efficiency of heat pump systems. Real-time monitoring of heat pump system performance necessitates continuously collecting and processing data from various components utilizing multiple sensors and controllers. This process is inherently complex and presents significant challenges. In recent years, artificial intelligence (AI) models have increasingly been applied in refrigeration, heat pump, and air conditioning systems due to their capability to identify and analyze complex patterns and data relationships, demonstrating higher accuracy and reduced computation time. In this study, multilayer perceptron (MLP), support vector machines (SVM), and random forest (RF) are used to develop COP prediction models for solar-assisted heat pumps. By comparing the predictive accuracy and modeling time of the three models built, the results demonstrate that the random forest model achieves the best prediction performance, with a mean absolute error (MAE) of 2.42% and a root mean squared error (RMSE) of 4.01% on the train set. On the test set, the MAE was 2.35% and the RMSE was 3.84%. The modeling time for the RF model was 6.57 s.

DeepRFQC: automating quality control for P-wave receiver function analysis using a U-net inspired network

This paper introduces DeepRFQC, an automated method for quality control in P-wave receiver function analysis. Leveraging a U-Net inspired deep learning model, which has previously shown promise in denoising and phase detection, DeepRFQC efficiently distinguishes usable from noisy receiver functions. We examine a Proterozoic Trans-Hudson Orogen dataset from northern Canada, including seismic events from 1990 to 2023, which is expanded for training purposes by data augmentation techniques. With 1,508,449 trainable parameters, the DeepRFQC model attains a commendable 96.6% validation accuracy, on a test dataset from the X5 seismic network; tests on stations from different tectonic environments indicate that the model is effective even in environments very different from the training set. Validation through the H-κ stacking method shows consistent and plausible results. As manual quality control is a major bottleneck in receiver-function processing, automated methods such as this one will allow for efficient examination of large data sets.

A Comprehensive Quality Evaluation for Gentiana Rigescens Franch. by Fingerprinting Combined with Chemometrics and Network Pharmacology.

Gentiana rigescens Franch. (G. rigescens) is a unique traditional medicinal herb from southwestern China, and its clinical mechanism for the treatment of hepatitis and the quality differences between different origins are not clear. The research aims to analyze the mechanisms for the treatment of hepatitis and differences in inter-origin differences using analytical techniques, chemometrics, and network pharmacology. Through infrared spectroscopy, spectral images, and high-performance liquid chromatography (HPLC) analysis, it was found that there were differences in absorbance intensity and significant differences in compound content among the samples' origin. G. rigescens iridoids and flavonoids exert therapeutic effects on hepatitis through multiple targets (GAPDH, EGFR, and MMP9, etc.) and multiple pathways (non-small cell lung cancer, hepatitis C, etc.). The above HPLC, chemometrics, and network pharmacology results revealed that gentiopicroside, and swertiamarine was the best quality marker among origins. The accuracy of the ResNet model train, test, and external validation sets for synchronous spectral images were 100 %, which could be utilized as an effective tool for tracing G. rigescens's origins. The R2 of the calibration and validation sets of the PLSR model was higher than 0.70. This model had excellent predictive performance in determining the content of gentiopicroside and swertiamarine, and could quickly, accurately, and effectively predict these two compounds. The research investigates the differences in G. rigescens origins from multiple perspectives, establishes image recognition models and prediction models, and provides new methods and theoretical basis for quality control of G. rigescens.

Generalization Limits of Data-Driven Turbulence Models

AbstractMany industrial applications require turbulent closure models that yield accurate predictions across a wide spectrum of flow regimes. In this study, we investigate how data-driven augmentations of popular eddy viscosity models affect their generalization properties. We perform a systematic generalization study with a particular closure model that was trained for a single flow regime. We systematically increase the complexity of the test cases up to an industrial application governed by a multitude of flow patterns and thereby demonstrate that tailoring a model to a specific flow phenomenon decreases its generalization capability. In fact, the accuracy gain in regions that the model was explicitly calibrated for is smaller than the loss elsewhere. We furthermore show that extrapolation or, generally, a lack of training samples with a similar feature vector is not the main reason for generalization errors. There is actually only a weak correlation. Accordingly, generalization errors are probably due to a data-mismatch, i.e., a systematic difference in the mappings from the model inputs to the required responses. More diverse training sets unlikely provide a remedy due to the strict stability requirements emerging from the ill-conditioned RANS equations. The universality of data-driven eddy viscosity models with variable coefficients is, therefore, inherently limited.

Self-learning salp swarm algorithm for global optimization and its application in multi-layer perceptron model training

Optimization problems are common across various fields, and one effective solution is the swarm intelligence algorithm.It is essential for the algorithm to deliver high-quality solutions for problems with varying characteristics. However, most existing swarm intelligence rely on fixed and monotonic search strategies, which limits their ability to handle the diverse and complex situations encountered when solving real-world optimization problems with unknown fitness landscapes. To extend the applicability of swarm intelligence and thus offer users an efficient black-box optimizer for various applications, a novel self-learning mechanism is proposed and applied to the Salp Swarm Algorithm (SSA) to develop the self-learning salp swarm algorithm (SLSSA) in this paper. In SLSSA, four distinct search strategies, including a novel multiple food sources search strategy, are adopted to strengthen the search agents’ abilities to conquer various difficulties in the search space. To improve the efficiency of the search process, the self-learning strategy dynamically determines the execution probability of each search strategy according to the quality of solutions it produced previously. Moreover, a parameter setting method is proposed in this paper, which eliminates the need for a trial-and-error approach and allows for straightforward configuration of the parameters that optimize the performance of SLSSA. In comparison with several highly regarded state-of-the-art peer algorithms, the performance of SLSSA in solving the CEC2014 benchmark functions was thoroughly examined. Subsequently, SLSSA was applied to train multi-layer perceptron classifiers and test on the UCI machine-learning datasets. The experimental results and analysis on benchmark functions and multi-layer perceptron classifier training problems demonstrate that SLSSA outperforms the competing algorithms in terms of solution accuracy, stability, and overall convergence speed. Moreover, computational time comparisons reveal that SLSSA achieves significant performance improvement with only a marginal increase in time cost compared to the original SSA.

Prognostic value of neutrophil to lymphocyte ratio and lymphocyte counts before durvalumab consolidation after radio-chemotherapy in locally advanced non-small cell lung cancer

BackgroundDurvalumab, an anti-PD-L1 immune checkpoint inhibitor, after radio-chemotherapy (RCT) has changed the management of locally advanced non-small cell lung cancer (LA NSCLC). A series of retrospective studies have investigated different cut-off of lymphocyte count (LyC) and neutrophil-to-lymphocyte ratio (NLR) to predict survival in LA NSCLC. None of these studies has validated their threshold in an independent group of patients. We wanted to assess the OS prognostic value of NLR and LyC in patients with LA NSCLC treated by RCT and durvalumab, with threshold determination and their validation in an external cohort.MethodsPatients were enrolled in four institutions between Oct. 2017 and Jan. 2022. Pre durvalumab LyC, neutrophils count (NC) and NLR were collected. To define NLR and LyC cut-off value predicting survival event, time dependent Receiver Operating Characteristics (ROC) curves was performed. Survival outcomes were estimated by the Kaplan-Meier method and differences were compared using univariate and multivariate Cox proportional hazard models.ResultsWe included 76 patients in the training set and 85 in the test set. The best cut off were 2,94 for NLR and 0,61 G/l for LyC to predict OS in the training set. For patients with NLR > 2,94, univariate analysis showed no significant deterioration in OS in either the training set (p = 0,066) or the test set (p = 0,12). Patients with LyC > 0,61 G/L, in univariate analysis, had longer OS in training set (p = 0,030) and in test set (p = 0,0062). This OS increase was not found in multivariate analysis (p = 0,057) in training set but was confirmed in test set (0,039).ConclusionLyC > 0,61 G/l is associated with longer OS for LA NSCLC patient’s treated with RCT and durvalumab in univariate analysis. In this context, a particular expectation for organs at risk sparing during RT to avoid lymphopenia seems important.Trial registrationRetrospectively registered.

Development and validation of a nomogram to predict the survival and estimate surgical benefits for gastric cancer with liver metastasis receiving primary tumor resection

BackgroundSurgical treatment has been widely controversial for gastric cancer accompanied by liver metastasis (GCLM). This paper aims to develop and validate a nomogram to predict the survival and estimate surgical benefits for GCLM patients.MethodsA total of 616 GCLM patients from the Surveillance, Epidemiology, and End Results Program (SEER) database and 74 GCLM patients receiving primary tumor resection (PTR) from the Chinese center were included in this study. Patients from the SEER database were divided into training set (with PTR) (n=493) and non-operative set (without PTR) (n=123). Patients undergoing PTR from China were included as external validation set. Independent risk factors associated with the overall survival of GCLM patients undergoing PTR were identified in the training set via log-rank test and Cox regression analysis. Afterwards, a comprehensive model and corresponding nomogram were constructed and validated by validation set.ResultsThe survival of patients undergoing PTR (n=493) was longer than that without PTR (n=123) (log-rank test, p<0.0001) in SEER cohort. T stage (HR=1.40, 95% CI=1.14, 1.73), differentiation grade (HR=1.47, 95% CI=1.17, 1.85), non-hepatic metastases (HR=1.69, 95% CI=1.29, 2.21), and adjuvant therapy (HR=0.34, 95% CI= 0.28, 0.42) were closely related with the survival of GCLM with PTR, and thus, a four-factor nomogram was established. However, GCLM patients receiving PTR in the high-risk subgroup (n=255) screened out by the nomogram did not have better survival outcomes compared with patients without PTR (n=123) (log-rank test, p=0.25).ConclusionsThe nomogram could predict survival of GCLM patients receiving PTR with acceptable accuracy. In addition, although PTR did improve the survival of whole GCLM patients, patients in the high-risk subgroup were unable to benefit from PTR, which could assist clinicians to make decisions for the treatment of GCLM.

An explainable deep learning model to predict partial anomalous pulmonary venous connection for patients with atrial septal defect

BackgroundPatients with partial anomalous pulmonary venous connection (PAPVC) usually present asymptomatic and accompanied by intricate anatomical types, which results in missed diagnosis from atrial septal defect (ASD). The present study aimed to explore the predictive variables of PAPVC from patients with ASD and constructed an explainable prediction model based on deep learning.MethodsThe retrospective study included 834 inpatients with ASD in Women and Children's Hospital, Qingdao University from January 2018 to January 2023. They were separated into two groups based on the presence of PAPVC. Propensity score matching and SMOTE were used to balance the baseline data between groups. The differential variables between the two groups were determined by univariate logistic regression. The patients were randomly divided into the training set and the validation set in a ratio of 8:2. Support vector machines (SVM), Random forest, Decision tree, XGBoost, and LightGBM were used to build models by differential variables. The classification performance of models was compared. Split, gain and SHAP were used to measure the importance of differential variables and improve the interpretability of the model. Moreover, a portion of the patients was included in the validation set to test the performance of the selected models.ResultsThree hundred twenty-eight patients with ASD and patients with 82 PAPVC were included in the training set and the validation set, respectively. The selection of 10 differential variables was based on univariate logistic regression, including right atrial diameter (longitudinal axis and transverse axis), right ventricular diameter, left atrial diameter, left ventricular end-diastolic diameter, left ventricular end-systolic diameter, P-wave voltage, P-wave interval PR interval, and QRS-wave voltage. In the classification model established based on differential variables, the LightGBM model achieved the highest performance on the validation set (AUC = 0.93). Based on variables importance analysis, the LightGBM-Clinic model was retrained by P-wave voltage, P-wave interval, PR interval, QRS wave interval, and right ventricular diameter, and performed excellently (AUC = 0.90). The AUC of the LightGBM-Clinic model was 0.87 in the test set.ConclusionIn this study, the LightGBM model performs excellently in determining whether patients with ASD are accompanied by PAPVC. ECG parameters such as P-wave voltage were important to predictive value and enhance the explainability of the model.

Cell Painting and Chemical Structure Read-Across Can Complement Each Other for Rat Acute Oral Toxicity Prediction in Chemical Early Derisking.

Early derisking decisions in the development of new chemical compounds enable the identification of novel chemical candidates with improved safety profiles. In vivo studies are traditionally conducted in the early assessment of acute oral toxicity of crop protection products to avoid compounds, which are considered "very acutely toxic", with an in vivo lethal dose of 50% (LD50) ≤ 60 mg/kg body weight. Those studies are lengthy and costly and raise ethical concerns, catalyzing the use of nonanimal alternatives. The objective of our analysis was to assess the predictive efficacy of read-across approaches for acute oral toxicity in rats, comparing the use of chemical structure information, in vitro biological data derived from the Cell Painting profiling assay on U2OS cells, or the combination of both. Our findings indicate that the classification of compounds as very acute oral toxic (LD50 ≤ 60 mg/kg) or not is possible using a read-across approach, with chemical structure information, morphological profiles, or a combination of both. When classifying compounds structurally similar to those in the training set, the chemical structure was more predictive (balanced accuracy of 0.82). Conversely, when the compounds to be classified were structurally different from those in the training set, the morphological profiles were more predictive (balanced accuracy of 0.72). Combining the two models allowed for the classification of compounds structurally similar to those in the training set to slightly improve the predictions (balanced accuracy of 0.85).

StructuGraphRAG: Structured Document-Informed Knowledge Graphs for Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating external data sources beyond their training sets and querying predefined knowledge bases to generate accurate, context-rich responses. Most RAG implementations use vector similarity searches, but the effectiveness of this approach and the representation of knowledge bases remain underexplored. Emerging research suggests knowledge graphs as a promising solution. Therefore, this paper presents StructuGraphRAG, which leverages document structures to inform the extraction process and constructs knowledge graphs to enhance RAG for social science research, specifically using NSDUH datasets. Our method parses document structures to extract entities and relationships, constructing comprehensive and relevant knowledge graphs. Experimental results show that StructuGraphRAG outperforms traditional RAG methods in accuracy, comprehensiveness, and contextual relevance. This approach provides a robust tool for social science researchers, facilitating precise analysis of social determinants of health and justice, and underscores the potential of structured document-informed knowledge graph construction in AI and social science research.